Molecular Diagnostics and Genetics Quantitative Analysis of Circulating Methylated DNA as a Biomarker for Hepatocellular Carcinoma

BACKGROUND: Hypermethylation of the RASSF1A [Ras association (RalGDS/AF-6) domain family member 1A] gene is frequently observed in hepatocellular car-cinoma (HCC). We evaluated the analysis of circulating hypermethylated RASSF1A for detecting HCC and assessing its prognosis.

RASSF1A 9 [Ras association (RalGDS/AF-6) domain family member 1A] as a model has demonstrated good diagnostic sensitivity and specificity for detecting placentally derived hypermethylated RASSF1A sequences in the plasma of pregnant women for the purpose of noninvasive prenatal diagnosis (11 ).In this study, we evaluated the clinical usefulness of this test for HCC detection and prognosis.
In the first module, we compared the serum concentrations of methylated RASSF1A sequences in HCC patients and 2 control groups (sex-and age-matched chronic HBV carriers and healthy volunteers) to evaluate the diagnostic sensitivity and specificity of this test.We also evaluated the prognostic value of methylated RASSF1A concentrations in the serum by conducting assays at several times after tumor resection.In the second module, we investigated the potential application of this test for HCC screening by measuring methylated RASSF1A concentrations in serum samples from 22 chronic HBV carriers who had undergone surveillance for HCC and had subsequently developed HCC.

STUDY PARTICIPANTS
Module 1. Sixty-three HCC patients undergoing surgical resection of HCC were recruited from the Department of Surgery, Prince of Wales Hospital.Three blood samples were collected from each HCC patient, at the time of diagnosis and at 1 month and 1 year after the surgery.The ethics approval committee of the institution approved the study.As a control, we recruited 1 chronic HBV carrier of the same sex and age for each HCC patient.We followed up all control individuals for at least 18 months, and none showed evidence of HCC.Table 1 summarizes the demographic information of the study participants.The high HBV prevalence in the HCC group is consistent with the fact that HBV infection is the most important etiologic factor for HCC in China (12 ).In contrast, the incidence of hepatitis C infection is very low.Fifty healthy individuals with no history of chronic hepatitis (median age, 48 years; 56% male) were recruited as another control group.
Module 2. A surveillance program for screening HCC in HBV carriers has been carried out at the Prince of Wales Hospital, Hong Kong.One thousand eighteen HBV carriers attending the hepatology clinic be-tween 1997 and 2000 were screened for HCC by AFP measurement and abdominal ultrasonography (13 ).Twenty-three carriers developed HCC during this surveillance period, and serum samples were available from 22 of them.As controls, we recruited 22 sex-and age-matched individuals who had enrolled in the surveillance program but who did not develop HCC.The median age of both groups was 56 years (interquartile range, 45-62 years), and 19 (86%) of these controls were male.The details of the surveillance program have been described elsewhere (13 ).

SAMPLE PROCESSING
HCC tissues and blood cells.To verify the specificity of the methylation-sensitive restriction enzymemediated real-time PCR system that we used to detect methylated DNA sequences, we used this detection system and bisulfite sequencing to analyze the methylation status of the RASSF1A and ACTB (actin, beta) promoters in tumor tissues and blood cells from 5 HCC patients.
Serum samples.DNA was extracted from 800 L of serum with the QIAamp DNA Mini Kit (Qiagen) as previously described (14,15 ).

RESTRICTION ENZYME
We digested 100 ng of DNA from HCC tissues and blood cells or 35 L of extracted serum DNA with 100 U of BstUI, a methylation-sensitive restriction enzyme, at 60 °C for 16 h.The methylation-sensitive enzyme degrades unmethylated DNA sequences, whereas methylated DNA sequences remain intact and detectable by real-time PCR.

ACTB SEQUENCES
We used a duplex real-time PCR assay to detect RASSF1A and ACTB sequences simultaneously (11 ).The sequences of the forward and reverse primers and the probe for RASSF1A are 5Ј-AGCCTGAGCT CATTGAGCTG-3Ј, 5Ј-ACCAGCTGCCGTGTGG-3Ј, and 5ЈFAM™-CCAACGCGCTGCGCAT(MGB™)-3Ј, respectively.The corresponding primer and probe sequences for ACTB are 5Ј-GCGCCGTTCCGAAAGTT-3Ј, 5Ј-CGGCGGATCGGCAAA-3Ј, and 5ЈVIC™-ACCG CCGAGACCGCGTC(MGB™)-3Ј, respectively.FAM and VIC are proprietary fluorescent dyes (Applied Biosystems), and MGB is a proprietary quencher (Applied Biosystems).Each reaction contained 1ϫ TaqMan Universal PCR Master Mix (Applied Biosystems), 300 nmol/L of each RASSF1A primer, 85 nmol/L of the RASSF1A probe, 450 nmol/L of each ACTB primer, and 126 nmol/L of the ACTB probe.We used 5 L of DNA not treated with enzyme or 7.15 L of an enzyme-digested serum DNA mixture (equivalent to 5 L of undigested serum DNA) as the template for each reaction.The thermal profile was 50 °C for 2 min, 95 °C for 10 min, and 50 cycles of 95 °C for 15 s and 60 °C for 1 min (11 ).We ran all reactions in duplicate and calculated the mean quantity.A DNA construct containing 1 copy each of the RASSF1A and ACTB amplicons was established as the quantitative calibrator.

PROMOTERS IN TUMOR TISSUES AND BLOOD CELLS
We treated 1 g of DNA with bisulfite, amplified the RASSF1A and ACTB promoters, and then cloned and sequenced the PCR products.CpG sites with cytosine residues sequenced as cytosine or thymine were scored as methylated or unmethylated, respectively.

PRECISION OF THE ANALYSIS OF SERUM METHYLATED RASSF1A
To investigate the reproducibility of this test for quantifying the circulating concentration of methylated RASSF1A, we evaluated the CV for the methylated RASSF1A assay.We digested 20 aliquots of a DNA sample extracted from an HCC tumor tissue with BstUI enzyme and quantified the methylated RASSF1A concentration by real-time PCR.The CV of the 20 replicate analyses was 11%.

STATISTICAL ANALYSIS
We compared the serum RASSF1A concentrations of HCC patients and matched HBV carriers with the Wilcoxon signed rank test and multiple logistic regression analysis, and we used the log-rank test and Cox proportional hazards regression analysis for survival analyses.Classification and regression tree analysis was used to evaluate RASSF1A cutoff values for combining the use of the AFP marker and RASSF1A analysis.P values Ͻ0.05 were considered statistically significant.

CONCORDANT RESULTS FOR ENZYME-MEDIATED REAL-TIME ANALYSIS AND BISULFITE SEQUENCING
To evaluate the specificity of this new system for detecting hypermethylated DNA sequences, we analyzed tumor tissues and blood cells from 5 HCC patients with bisulfite sequencing and with the enzyme-mediated real-time detection system.In the bisulfite-sequencing analysis, we detected RASSF1A promoter hypermethylation in all tumor tissues but in none of the blood cell samples.In contrast, the ACTB promoter was completely unmethylated in both sample types.The results of the enzyme-digestion analysis were in complete accord with the bisulfite-sequencing results.After enzyme digestion, RASSF1A sequences were detectable only in the tumor tissues, whereas ACTB sequences were detectable in neither tumor tissues nor blood cells.Fig. 1 shows the results for bisulfite sequencing and real-time PCR analysis for 1 case.

DETECTION OF SERUM ACTB SEQUENCES AFTER ENZYME DIGESTION
We incorporated a system in our analysis to check the completeness of enzyme digestion to minimize the chance of false-positive results caused by incomplete digestion of the background unmethylated RASSF1A sequences.This checking system consisted of a PCR assay that amplified the ACTB promoter with the same number of restriction sites as in the RASSF1A amplicon.Because we have previously shown that the digestion efficiencies for RASSF1A and ACTB sequences are similar (11 ), the presence of an ACTB signal in a sample suggests that the digestion of unmethylated RASSF1A is incomplete.In such cases, the DNA samples would be redigested for 8 h.

DETECTION OF RASSF1A SEQUENCE AFTER DIGESTION OF SERUM SAMPLES FROM HCC PATIENTS
In module 1, RASSF1A sequences were detectable after enzyme digestion in the sera of 93% (59 of 63) of the HCC patients at diagnosis and in 58% (37 of 63) of the HBV carriers.The median concentrations for the 2 groups were 7.70 ϫ 10 5 copies/L and 1.18 ϫ 10 5 copies/L, respectively (P Ͻ 0.0001, Wilcoxon test).The difference between the 2 groups in the total serum concentration of RASSF1A (without enzyme digestion) was not statistically significant (P ϭ 0.5, Wilcoxon test).Only 4 (8%) of the 50 healthy individuals showed detectable RASSF1A after digestion.The median concentration for these 4 individuals was 3.7 ϫ 10 4 copies/L (Fig. 2A).The area under the ROC curve for dis-tinguishing between HCC patients and HBV carriers was 0.81.
In contrast, serum RASSF1A concentrations for HCC patients with and without ultrasonographic evidence of cirrhosis were not statistically different (median, 7.98 ϫ 10 5 copies/L vs 7.00 ϫ 10 5 copies/L; P ϭ 0.529, Mann-Whitney U-test; Fig. 2B).The difference in serum RASSF1A concentrations between HCC patients and HBV carriers was independent of the presence of ultrasonographic and biochemical changes due to cirrhosis (logistic regression; see Table 1).

SERUM RASSF1A CONCENTRATIONS AFTER SURGICAL RESECTION
Of the 59 patients with detectable methylated RASSF1A in the serum at diagnosis, 45 patients (76%) showed a reduced concentration 1 month after tumor resection.

DISEASE-FREE SURVIVAL
Patients with serum RASSF1A concentrations greater than the median concentration of the group at diagnosis showed significantly poorer disease-free survival (defined as survival without any clinical evidence of disease recurrence) than the patients with lower concentrations (P ϭ 0.0034, log-rank test; Fig. 3A).The difference in survival probabilities was independent of the presence of biochemical and ultrasonographic evidence of cirrhosis (P ϭ 0.0028, Cox proportional hazards regression).Similarly, the RASSF1A concentration 1 year after surgery was also associated with survival (P ϭ 0.031, log-rank test; Fig. 3C).On the other hand, the concentration 1 month after the operation was not associated with survival (P ϭ 0.33, logrank test; Fig. 3B).

CORRELATION BETWEEN THE SERUM CONCENTRATIONS OF AFP AND RASSF1A
The serum concentrations of RASSF1A and AFP were not significantly correlated, either for the HCC patients (P ϭ 0.5, Spearman correlation analysis) or for the chronic HBV carriers (P ϭ 0.91, Spearman correlation analysis), suggesting that concurrent use of both markers might provide a nonoverlapping and potentially synergistic set of information.The cutoff values for methylated RASSF1A in serum were determined by classification and regression tree analysis for classifying individuals with typical or increased AFP concentrations.For individuals with increased AFP, we used a lower RASSF1A cutoff value of 1.50 ϫ 10 5 copies/L to screen out false-positive AFP results, and we used a higher RASSF1A cutoff value of 1 ϫ 10 6 copies/L for individuals with typical AFP concentrations to identify HCC patients with false-negative AFP results.Twentytwo (35%) of the 63 HCC patients in module 1 had typical AFP concentrations of Ͻ20 g/L.A serum RASSF1A cutoff value of 1 ϫ 10 6 copies/L further identified 11 patients (50%) who had false-negative AFP results (Fig. 4).On the other hand, 4 of the 8 HBV carriers with increased AFP values had RASSF1A concentrations of Ͻ1.50 ϫ 10 5 copies/L.

SERUM RASSF1A CONCENTRATIONS IN HBV CARRIERS UNDER SURVEILLANCE FOR HCC
In module 2, we studied 22 HBV carriers who developed HCC during the HCC screening program.The Brackets and P values refer to the indicated group comparisons.The upper, middle, and lower lines of each box represent the 75th, 50th, and 25th percentiles, respectively.The upper and lower whiskers represent the 95th and 5th percentiles.

Fig. 3. Disease-free survival probabilities in HCC patients according to the serum RASSF1A concentration after enzyme digestion at diagnosis (A) and at 1 month (B) and 1 year (C) after tumor resection.
Disease-free survival is defined as survival with no clinical evidence of disease recurrence.For each time point, the patients were divided into 2 groups with the median concentration for all patients as a cutoff value.The median concentrations at diagnosis and at 1 month and 1 year after the operation were 7.70 ϫ 10 5 , 2.50 ϫ 10 5 , and 9.10 ϫ 10 5 copies/L, respectively.For the survival analysis with the serum concentration at 1 year, only the patients with no clinical evidence of residual disease at 1 year were analyzed.
median time between enrollment and HCC diagnosis was 30 months, and we observed a significant increase in serum RASSF1A concentration from enrollment to HCC diagnosis (median, 3.74 ϫ 10 5 copies/L to 5.05 ϫ 10 5 copies/L; P ϭ 0.014, Wilcoxon test; Fig. 5).For the matched control individuals who had not developed HCC, the serum RASSF1A concentrations at the 2 cor-responding time points were not significantly different (median, 1.16 ϫ 10 5 copies/L vs 0.92 ϫ 10 5 copies/L; P ϭ 0.495).At the second time point, the patients who had developed HCC had significantly higher methylated RASSF1A concentrations in the serum than the control individuals (P Ͻ 0.001, Wilcoxon test).Twelve (60%) of the 22 patients who had developed HCC were classified as having HCC with the combined AFP and RASSF1A algorithm (Fig. 4) at the time of HCC confirmation.In comparison, AFP measurement alone identified only 8 cases (40%).Moreover, only 1 of the 22 control individuals fulfilled the combined criteria for HCC at the time of enrollment; none fulfilled these criteria at follow-up.

Discussion
Hypermethylation of the RASSF1A promoter is frequently observed in HCC and has been shown to play an important role in HCC pathogenesis (9,16,17 ); however, methylation-specific PCR detects hypermethylated RASSF1A sequences in the circulation of only 40%-70% of HCC patients (18,19 ).These low detection rates are probably related to the substantial degradation of DNA (up to 96%) caused by the bisulfite conversion step used in methylation-specific PCR (20 ) and the low concentration of circulating tumor DNA in cancer patients (21,22 ).We recently developed a nonbisulfite method for detecting and quantifying circulating hypermethylated RASSF1A sequences (11 ).With this method, we detected methylated RASSF1A sequences in 93% of HCC patients.This frequency is similar to that previously reported for detecting RASSF1A hypermethylation in HCC tumor tissues (9,18 ).The substantial improvement is likely to be related to specific degradation of unmethylated sequence by the methylation-sensitive restriction enzyme, in contrast to the nondiscriminatory degradation of both methylated and unmethylated DNA with bisulfite conversion (20 ).This improvement also has enhanced the reliability of the quantitative measurement.Thus, our study is the first to show that the concentration of circulating hypermethylated RASSF1A is associated with disease-free survival in HCC patients after tumor resection.
Despite the extensive use of methylation-sensitive restriction enzymes in epigenetic research, methylation-sensitive restriction enzyme analysis has not been popular for detecting aberrant DNA methylation in the circulation for the purpose of cancer detection.The reason is probably related to the difficulties in differentiating false-positive results due to incomplete enzyme digestion from the presence of a low concentration of tumor-derived hypermethylated sequences.In this regard, we incorporated several measures to mini-  mize the potential problem of incomplete digestion.These measures included (a) the inclusion of multiple enzyme restriction sites in the PCR amplicon, (b) the use of excess restriction enzyme and a prolonged incubation period, and (c) the introduction of a system to check the completeness of enzyme digestion.Our check for incomplete digestion demonstrated Ͼ10% of the samples are potentially incompletely digested after the initial enzyme-digestion step.
The presence of low concentrations of methylated RASSF1A in the sera of 58% of the chronic HBV carriers is consistent with previous reports that RASSF1A promoter hypermethylation is an early event in HCC pathogenesis and can be observed in nonmalignant liver tissues of patients with cirrhosis or chronic hepatitis (16,23 ).On the other hand, the predominant source of circulating methylated RASSF1A in HCC patients is likely the tumor cells for the following reasons.First, the median concentration of methylated RASSF1A in the serum was Ͼ6-fold higher in HCC patients than in matched HBV carriers, a difference that was independent of the presence of cirrhosis.Second, a significant reduction in serum RASSF1A was observed in the HCC patients after tumor resection.Lastly, we observed a significant increase in the circulating concentration of methylated RASSF1A from the time of recruitment to the time of HCC development in the HBV carriers who subsequently developed HCC.
Diagnostically, the absence of a correlation between the serum concentrations of AFP and RASSF1A has allowed the synergistic use of these 2 biomarkers to improve their diagnostic accuracies.The use of classification and regression tree analysis has produced a different serum RASSF1A cutoff values for individuals with increased AFP concentrations than for those without increased AFP concentrations.The diagnostic sensitivity and specificity were 77% and 89%, respectively, for this combined AFP and RASSF1A analysis, compared with 65% and 87%, respectively, for AFP measurement alone.This improved diagnostic accuracy may be useful for early detection of HCC, because increased AFP concentrations have been shown in only about 40% of HCC patients with early-stage disease (8 ).
In addition to HCC detection, analysis of methylated RASSF1A in the serum also has prognostic value in that the concentrations in these patients at diagnosis and 1 year after surgery are associated with disease-free survival.This prognostic value was also independent of the presence of ultrasonographic and biochemical evidence of cirrhosis.These findings suggest that methylated RASSF1A in the serum reflects the tumor load in HCC patients.A correlation between circulating tumor DNA and tumor load has also been described for other cancers (24,25 ).The apparent lack of a correlation between the serum concentration of RASSF1A at 1 month after the operation and survival may be related to the background concentration of methylated RASSF1A derived from nonmalignant hepatitic liver tissues.In the presence of a tumor, a high proportion of the circulating RASSF1A would be derived from tumor cells, and hence the RASSF1A concentration could reflect the tumor load.After tumor resection, however, the methylated RASSF1A sequences released from nonmalignant hepatitic liver tissue may obscure the small amounts of RASSF1A sequences derived from residual cancer cells.Because the RASSF1A concentration observed at 1 year after tumor resection (but not the concentration at 1 month) was associated with disease recurrence, monitoring the changes in RASSF1A may be useful for identifying disease recurrence.
In this study, we demonstrated the clinical usefulness of the methylation-sensitive restriction enzymemediated PCR approach for detecting aberrantly methylated tumor suppressor gene sequences in the circulation of cancer patients.This method permitted the detection of methylated RASSF1A sequences in the sera of 93% of HCC patients.This detection rate is better than that reported for bisulfite-based detection methods.Adopting different cutoff values for individuals with typical and increased AFP concentrations may facilitate the use of serum RASSF1A analysis for HCC detection.Although hypermethylation of the RASSF1A promoter is not specific for HCC (26 -29 ), in the context of HCC screening in a chronic hepatitis carrier, an observed increase in the circulating concentration of methylated RASSF1A would definitely warrant further investigation for HCC, as via an imaging study for example.The presence of very low concentrations of methylated RASSF1A in the healthy control individuals is an interesting finding.Previous studies have shown that hypermethylation of the RASSF1A promoter may be present in a very low percentage of premalignant lesions (30 ) and even nonpathologic tissues (31 ).Diagnostically, none of these healthy individuals had circulating concentrations of methylated RASSF1A greater than the cutoff of 1.50 ϫ 10 5 copies/L proposed for HCC diagnosis.Technically, this method is relatively simple and inexpensive and hence has the potential to be adapted for routine clinical use.More importantly, the application of this technology would not be limited to detecting aberrantly methylated RASSF1A for HCC detection only.Because aberrant methylation of tumor suppressor genes has been extensively reported for virtually all cancers, this technology can be generalized for the detection of different types of cancer via the use of different panels of tumor suppressor genes (26 -29 ).
Grant/Funding Support: This work has been supported by the Innovation and Technology Commission of the Hong Kong SAR Government under the Innovation and Technology Support Program (ITS/014/06).Financial Disclosures: The authors have filed a patent application on the technology described in this report.

Fig. 1 .
Fig. 1.Bisulfite-sequencing (left) and the corresponding real-time amplification analysis (right) results of a representative HCC patient.For the bisulfite-sequencing analysis, each row represents 1 DNA molecule.The CpG sites are numbered according to their sequence in the PCR amplicon of the particular assay.The first CG dinucleotide in the PCR amplicon is numbered as 1.The CpG sites within recognition sites of the methylation-sensitive restriction enzyme are underlined.Open and closed circles represent unmethylated and methylated CpG sites, respectively.

Fig. 2 .
Fig. 2. Serum concentrations of RASSF1A sequences after digestion in HCC patients, HBV carriers, and healthy individuals (A) and in HCC patients and HBV carriers with and without ultrasonographic evidence of cirrhosis (B).

Fig. 5 .
Fig. 5. Serum RASSF1A concentrations in chronic HBV carriers undergoing HCC surveillance who did and did not develop HCC.For the group with HCC development, serum RASSF1A concentrations were measured at the time of enrollment and when HCC was confirmed.For the group without HCC development, the second measurement was taken as close as possible to the time of HCC diagnosis for the corresponding HCC patient.

Table 1 . Demographics of the HCC patients and the sex-and age-matched HBV carriers (N ‫؍‬ 63).
a Data are presented as the median (interquartile range).b NA, not applicable; HCV, hepatitis C virus.c Wilcoxon signed rank test.d 2 test.e Bilirubin, albumin, and prothrombin time are biochemical changes used for grading the severity of cirrhosis in the Child-Pugh staging system.